The Term Structure of Asset Return Informativeness

What is Information?

Term Structure of Asset Return Informativeness

• Prices move when news hits the market — but do they reflect information?
• If prices revert quickly, does the initial price movement matter?
• If prices underreact, is the initial price movement a good measure of informativeness?

This paper

How much of an event-day return persists and conveys information about future prices?

• We introduce a new approach, relative informativeness, to measure how informative return reactions are — across time horizons.

• We use this approach to study:
  • Earnings announcements
    
  • Intraday vs overnight returns
    
  • Macroeconomic news

Results Preview

Related Literature: Return Informativeness

• Informativeness
  • Bai, Philippon, and Savov (2016); Dávila and Parlatore (2024)
• Unbiasedness and price discovery
  • Biais, Hillion, and Spatt (1999); Barclay and Hendershott (2003): Learning in trading
    
  • Van Kervel and Menkveld (2019): HFTs and intraday price informativeness
    
  • Akey, Grégoire, and Martineau (2022): Insider trading and price discovery
    
  • Boguth et al. (2023): How news affects learning from prices
    
  • We extend this work by estimating how much of a return reflects informational content at multiple horizons
• Attention and horizon bias
  • Dessaint, Foucault, and Fresard (2024): Short-term data improves near-term forecasts, worsens long-term
    
  • Farboodi et al. (2022): Attention biased toward large firms
    
  • Farboodi and Veldkamp (2020): Announcements increase uncertainty
    
  • Boguth et al. (2023), Hu et al. (2022): VIX response reflects informational risk

Methodology

gantt
    %% This is a comment
    dateFormat mm
    axisFormat %M
    Event return          :a1, 00, 1m
    Horizon return :00, 11m

We estimate how event-day returns relate to longer-horizon returns:

r_{h,i} = \alpha_{h,\tau} + \beta_{h,\tau} \, r_{\tau,i} + \epsilon_i

• r_{\tau,i}: return during event window \tau
• r_{h,i}: return over horizon window h, where h begins at and includes \tau

We care mostly about \beta and R^2:

• \alpha: background drift (i.e. average risk premium)
• \beta: fraction of event return that persists
• R^2: strength of the relationship

Under the Null: Random Walk

If returns follow a random walk:

• \beta_{h,\tau} = 1: full incorporation, no predictability
• R^2_{h,\tau} should follow:

E[R^2_{h,\tau}] = \frac{\text{length of } \tau}{\text{length of } h}

This gives us a theoretical baseline — but in practice we use empirical benchmarks from pseudo-events to better match the auto-correlation and heteroskedasticity of returns.

Relative Informativeness

We define:

\Psi_{h,\tau,e} = \frac{R^2_{h,\tau,\text{event}}}{R^2_{h,\tau,\text{benchmark}}}

• Numerator: R^2 from actual event days
  
• Denominator: R^2 from benchmark days with same horizon h

\Psi tells us how much more (>1) or less (<1) informative an event is than a typical day.

Benchmark

• \Psi is a relative measure
• Choice of benchmark depends on what we want to measure

Potential benchmarks

• All days: informativess relative to an average day
• Non-event days: informativess relative to an average non-event day day
• Alternate period: informativess relative to other period
  • e.g. overnight vs intraday
• Untreated firms: informativess relative to untreated firms
  • Useful for cross-sectional regressions
  • Can use matched control to create sample

Relationship to Forecasting Regressions

Suppose the horizon excludes the event window:

r_{h',i} = \alpha_{h',\tau} + \beta_{h',\tau} \, r_{\tau,i} + \epsilon_i

Then:

\beta_{h,\tau} = 1 + \beta_{h',\tau}

• \beta_{h,\tau} \neq 1 implies the event return contains forecastable information about future returns.

Simulations

We use simulations to understand how \Psi and \beta behave under different return-generating processes.

• Simulate 30 years of daily returns
• Events randomly spaced (on average every 21 days)
• Horizon lengths: 5 to 252 days
• Compare different models: random walk, persistent shocks, reversals, etc.

Baseline: Random Walk

Daily returns are i.i.d. normal, event shocks are indistinguishable from non-event days

Large Shock, No Persistence

Event return = 10× average shock, no continuation or reversal

Persistent Small Shocks

Event return = 0.5× average shock, but strongly persistent (AR(1) = 0.95)

Large Shock with Reversal

Large event return, but designed to fully reverse over ~21 days, mimics discount rate shock

Statistical Inference

We use two complementary inference methods:

1. Bootstrap inference: to construct confidence intervals for our estimates
   
2. Randomization inference: to tests whether event days are more informative than random benchmark days

Data

Sample period: 1994–2023 (plus 2024 returns for full 1-year horizons)

CRSP daily:

• S&P 500 ETF (SPY) - daily returns + open/close price
• U.S. Treasury ETFs (SHY, IEF, TLH, TLT)
• Individual CRSP stocks (for earnings)

Events:

• Earnings Announcements: I/B/E/S
• Macroeconomic News: Bloomberg
• Assigned to return window based on release timing

Related Literature : Earnings

• Savor and Wilson (2016): Systematic risk around earnings
  
• Beaver (1968): Earnings and return volatility
• Bernard and Thomas (1989); Martineau (2021): PEAD

Earnings Announcements: Market-Level Analysis

• SPY return behavior when >2% of S&P 500 firms report vs. typical days (18.3%)

Firm-Level: Informative and Persistent

• Sample: 2,089 CRSP firms with ≥10 years of data and ≥40 earnings announcements

Small vs Large Firms

• Partition firms into small and large (median decile over time)

Related Literature : Overnight Returns

• Cliff, Cooper, and Gulen (2008): Equity premium driven by overnight returns
  
• Hendershott, Livdan, and Rösch (2020): CAPM holds overnight
  
• Boyarchenko, Larsen, and Whelan (2023); Bondarenko and Muravyev (2023): Positive drift in futures
  
• Branch and Ma (2012); Berkman et al. (2012): Reversals intraday
  
• Our contribution: isolate informativeness from drift or noise

Overnight vs Intraday Returns

• Overnight: Close-to-open
• Intraday: Open-to-close

• Full sample: 1994–2024
  • Overnight returns contribute most of the mean daily return
    
  • Intraday returns have higher volatility and larger absolute moves

Overnight vs Intraday Informativeness

Overnight vs Intraday - By Decade

Overnight vs Intraday - Rolling 5-Year Estimates

What Changed?

• Rise of S&P E-mini futures (introduced 1997)
  • Electronic trading
  • Extended trading hours, especially post-2005.
• Reg NMS (2005) and algorithmic market structure
• Overnight trading became more active and informational

Related Literature: Macroeconomic News

• Macroeconomic news and price impact
  • Focus on immediate reactions:
    • Stocks (Bernanke and Kuttner 2005; Gilbert 2011)
      
    • FX (Anderson et al. 2003; Andersen et al. 2007)
      
    • Bonds (Fleming and Remolona 1999)
  • Persistence is limited:
    • Boguth et al. (2023): FOMC announcements reverse within 10 days
      
  • Our contribution: focus on term structure of informativeness

• Asset pricing around macro announcements
  • Lucca and Moench (2015): Pre-FOMC drift in equities
    
  • Chaudhry (2020): Event timing and asset responses
    
  • Kroencke, Schmeling, and Schrimpf (2021), Hu et al. (2022): FOMC-driven time-varying risk premia
    
  • Liu, Tang, and Zhou (2022), Cujean and Jaeger (2023): Structural modeling of FOMC risks
    
  • Our angle: measure how much lasting information these events contribute

Do Macro Announcements Matter?

• Macro events are widely watched
• Generate volatility spikes
• But does volatility = information?

We study the informativeness of five major types of macro announcements:

• FOMC announcements
  
• FOMC minutes
  
• CPI (inflation)
  
• GDP
  
• Unemployment

We assess informativeness for both equities (SPY) and Treasuries (various maturities).

Macro Announcements

S&P 500 - All macro announcements

S&P 500 - By Decade

Treasuries

Maturities vs. Horizons

Takeaways

• Macro news ≠ persistent signal
  
• Use of macro days in event studies must consider:
  • Horizon
  • Asset class
  • Type of announcement
• Treasury responses reflect fundamental repricing; equities often reflect noise, sentiment, or other “temporary” factors

Concluding Remarks

We introduce Relative Informativeness (\Psi): a measure of how much of an event day’s return reflects durable information. It’s applicable across event types, horizons, and asset classes, and focuses on explanatory power rather than simply price response.

Empirical Findings: Earnings are mean-reverting at the market level but informative firm-level; overnight reactions are now as strong as intraday (post-2010s); and macro news exhibits varying persistence.

Future work: Applying \Psi to all types of news and/or events:

• Earnings (more in depth)
• Corporate events
• Intraday price movements
• International news

Ultimately, use \Psi to ask: Are returns informative about future prices?

Thank you!